Projects

### Overview: tty7 is a terminal emulator that outlives its window, providing persistent sessions, remote work, and agent-driven functionality. It is built with pure Rust and GPU rendering, making it 2x faster than other similar terminals. ### Key features: - **Persistent Sessions**: Quits the app and your shells continue running; sessions resume after a reboot. - **Agent-Aware**: Supports Claude Code, Codex, and other agents, providing status, notifications, and context for each repository. - **Scriptable by Agents**: One agent can open a pane for another, execute tasks, and wait for results. - **Editor-Grade Input**: Suggestions, completion, highlighting, and history search without plugins. - **Remote Development**: Files, repos, panes, and git data remain on the remote machine, accessible via native SSH. ### Tech stack: Pure Rust, GPU rendering with Zed's gpui, VT core from Alacritty. ### Why it matters: - **Performance**: ~2x throughput compared to Alacritty, Ghostty, or Kitty. - **Persistent Sessions**: No need for tmux; sessions resume after a reboot. - **Agent-Aware**: Status, notifications, and context for each repository. - **Scriptable by Agents**: One agent can open a pane for another, execute tasks, and wait for results. - **Editor-Grade Input**: Suggestions, completion, highlighting, and history search without plugins. - **Remote Development**: Files, repos, panes, and git data remain on the remote machine, accessible via

### Overview **skill-up** is a CLI tool designed to evaluate Agent Skills, agents, and their workspaces. It evaluates responses and workspace changes, and produces reports locally or in CI. The tool runs declarative cases against an Agent Engine, grades the response and workspace changes, and uses **skill-upper** to inspect failures, repair or expand the eval suite, and rerun skill-up through conversation. ### Key Features - **Skill evaluation**: Measures a Skill's behavior across cases and can compare runs with and without the Skill. - **Agent evaluation**: Runs the same cases without installing a Skill, allowing assessment of an engine or custom agent. - **Workspace evaluation**: Checks how an agent works with files and repositories, using case fixtures or an existing local workspace. - **Skill evolution**: Uses **skill-upper** to inspect failures, repair or expand the eval suite, and rerun skill-up through conversation. ### Tech Stack Go ### Why it Matters Skill-up is a powerful tool for evaluating and evolving Agent Skills, agents, and workspaces. It provides a declarative approach to evaluation, supports multiple Agent Engines, and includes features like skill evolution, deep observation capture, and isolation runs. The bundled **skill-upper** plugin enhances the tool with features like approval-gated regression cases, isolated runs, and status comparison.

Overview: Duckling is a Clojure library designed to parse text into structured data, offering features such as parsing time, duration, and temperature expressions. Key features: - Parses text into structured data - Supports time, duration, and temperature expressions - Offers a simple API for easy integration into Clojure projects Tech stack: Clojure Why it matters: Duckling simplifies text parsing, making it easier to extract structured data from natural language text, enhancing the efficiency of applications that require accurate time, duration, and temperature calculations. Repository: facebookarchive/duckling_old GitHub stars: 1321 GitHub description: Deprecated in favor of https://github.com/facebook/duckling

Overview: ModelFusion is a TypeScript library designed for building AI applications, providing a unified API for common operations such as text streaming, object generation, and tool usage. Key features: - Supports text generation, image generation, vision, text-to-speech, speech-to-text, and embedding models. - Infers TypeScript types and validates model responses. - Provides an observer framework and logging support. - Ensures seamless operation through automatic retries, throttling, and error handling mechanisms. - Fully tree-shakeable and can be used in serverless environments. Tech stack: TypeScript Why it matters: ModelFusion simplifies the integration of AI models into JavaScript and TypeScript applications, making it easier to build AI applications, chatbots, and agents. It supports a wide range of models and provides robust observability and logging features, ensuring seamless operation and robustness.

### Overview: PyTorch-CycleGAN is a clean and readable implementation of the CycleGAN model, designed for applications in image-to-image translation tasks. It simplifies the original implementation by removing unnecessary complexity and providing a more straightforward interface. ### Key Features: - **Model Implementation**: A clean and readable PyTorch implementation of the CycleGAN model. - **Dataset Setup**: Supports the use of pre-existing datasets or custom datasets by following the provided directory structure. - **Training**: Automatically downloads and prepares datasets, trains the CycleGAN model, and saves weights. - **Testing**: Generates test outputs and allows customization of parameters. - **Visualization**: Uses Visdom to visualize training progress and loss graphs. ### Tech Stack: - **Python**: The primary programming language used for development. - **PyTorch**: The deep learning framework used for implementing the model. - **Visdom**: A library for visualizing training progress and loss graphs. ### Why It Matters: This project simplifies the implementation of CycleGAN, making it easier for developers to use and understand, especially for those new to the model or deep learning in general. The clean and readable codebase ensures better maintainability and reduces the learning curve for developers, making it a valuable resource for anyone working on image-to-image translation tasks.

### Overview: AI Runner is a private AI companion and art canvas built with privacy, flexibility, and creativity in mind. It runs on your machine and is designed for offline use, allowing you to shape your companion and create AI art without an internet connection. ### Key Features: - **🤖 AI Companion**: Shape a named, voiced companion with persistent personality, shifting mood, and long-term memory built from your conversations. - **🎨 Layered Canvas**: Draw, paint, generate, and filter on a multi-layer canvas — convert sketches to images, composite scenes, and iterate in place. - **🖼️ Image Generation**: Z-Image Turbo with embeddings, image-to-image, and more. ### Tech Stack: Python ### Why it Matters: AI Runner is a private AI companion and art canvas built with privacy, flexibility, and creativity in mind. It runs on your machine and is designed for offline use, allowing you to shape your companion and create AI art without an internet connection.

Overview: LangChat 是一个开源的 AI Agent 应用平台，支持多模型接入、Agent 构建、知识库 RAG、Skills、MCP 等功能，适用于企业级 AI 平台选型与采购。 Key features: - 多模型接入 - Agent 构建与调试 - 对话运行时支持 - 知识库 RAG - Skills 扩展 - MCP 集成 - 智能问数与图像能力 - 应用市场与权限管理 Tech stack: Java, Vue Why it matters: LangChat 为开发者提供了一个模块化、可扩展的平台，简化了 AI 应用的构建与部署，适用于企业级 AI 平台选型与采购。

Overview: Text-to-LoRA (T2L) is a reference implementation that adapts large language models (LLMs) for specific benchmark tasks using only textual task descriptions as input. Key features: - Hypernetworks that adapt LLMs - Textual input for task description - Efficient adaptation process - Works with various LLMs, including LLaMA and Gemma Tech stack: Python Why it matters: T2L simplifies the process of adapting LLMs for specific tasks, making it accessible for researchers and developers without requiring extensive expertise in model architecture or fine-tuning.

Overview: EmbedAnything is a highly performant, modular, and memory-safe Rust-based pipeline for generating embeddings from various sources, including text, images, audio, PDFs, and websites. It supports dense, sparse, ONNX, model2vec, and late-interaction embeddings, making it versatile for a wide range of use cases. Key features: - Streamline the process of generating embeddings from various sources. - Seamless streaming (memory-efficient indexing) of embeddings to a vector database. - Supports dense, sparse, ONNX, model2vec, and late-interaction embeddings. - Built with Rust for speed and efficiency. Tech stack: Rust. Why it matters: EmbedAnything is a minimalist yet highly performant solution for embedding generation, making it an essential tool for developers working with diverse data sources and vector databases.

Overview: This repository contains interview notes and answers for algorithm engineers specializing in large language models (LLMs). Key features: - Interview notes and answers - Algorithm engineering expertise - LLMs related Tech stack: Python, Jupyter Notebook Why it matters: This resource is valuable for algorithm engineers looking to prepare for and understand the technical aspects of working with large language models. Repository: jackaduma/awesome_LLMs_interview_notes GitHub stars: 1319

### Overview: Neo4j GraphRAG for Python is an official library that enables developers to build graph retrieval augmented generation (GraphRAG) applications using the power of Neo4j and Python. It offers a robust, feature-rich, and high-performance solution with long-term support from Neo4j. ### Key features: - GraphRAG application development - Knowledge Graph creation - Knowledge Graph enrichment with vector search - Knowledge Graph retrieval using Neo4j - Knowledge Graph creation pipelines ### Tech stack: Python ### Why it matters: This package provides a powerful and reliable solution for building GraphRAG applications, leveraging the capabilities of Neo4j and Python. It offers a robust, feature-rich, and high-performance solution with long-term support from Neo4j, making it an essential tool for developers working with Neo4j and knowledge graph applications.

Overview: NeuroAPI предоставляет безопасные, один-click установки для Codex CLI и Claude Code, доступные через российский AI API NeuroAPI. Это позволяет пользователям быстро настроить и использовать эти популярные инструменты для обработки текста и создания контента. Key features: - NeuroAPI: российский AI API для Codex CLI и Claude Code - One-click setup for Codex CLI and Claude Code - Secure and reliable installation - Supports Windows and macOS - Easy integration with Codex Desktop Tech stack: PowerShell Why it matters: NeuroAPI simplifies the process of integrating popular AI tools like Codex CLI and Claude Code with a secure and reliable setup, making it easier for developers to use these tools without the need for frequent updates or manual configuration.

### Overview: The COG-second-brain project is a self-evolving second brain powered by AI agents, markdown files, and version control. It includes 35 AI skills, 10 agents, and a people CRM, working with Claude Code, Cursor, Kiro, Gemini CLI, and OpenAI Codex. ### Key features: - **35 AI Skills**: Capture raw thoughts, classify news, save URLs, analyze cross-domain patterns, and more. - **10 Agents**: Support for Claude Code, Antigravity, Cursor, Kiro, Gemini CLI, and OpenAI Codex. - **People CRM**: Tracks interactions and insights for better understanding and management of users. ### Tech stack: The project uses Python for the core functionality and markdown files for storing and processing data. ### Why it matters: The self-evolving nature of COG allows it to continuously improve and adapt, making it a powerful tool for personal and professional productivity.

### Overview: SeaGOAT is a local-first semantic code search engine designed for the AI age. It leverages vector embeddings to enable semantic search, making it ideal for developers working with large codebases. ### Key Features: - **Semantic Search**: SeaGOAT searches codebase semantically, allowing developers to find relevant snippets based on context. - **Vector Embeddings**: Utilizes vector embeddings to represent code snippets, enabling precise and context-aware search. - **Local Server**: Runs as a local server, ensuring fast response times and no need for internet connection. - **Regular Expressions Support**: Supports regular expressions in queries, enhancing the flexibility of search. - **Configuration**: Configurable through YAML files, allowing for customization of the search engine. - **Generative Providers**: Supports generative search providers like Ollama, Minimax, and OpenAI, enhancing the search experience. ### Tech Stack: Python, Ripgrep, Bat ### Why it Matters: SeaGOAT is designed to help developers efficiently find and understand code within their projects, making it a valuable tool for modern software development. Its semantic search capabilities and local server functionality make it particularly useful in large-scale development environments. Repository: kantord/SeaGOAT GitHub Stars: 1313

**Overview:** Reverify is a tool designed to stop AI models from making unfounded claims by proposing, checking, and verifying every claim against ground truth. It uses a deterministic Python toolkit to ensure that only verified facts are accepted, making it particularly useful in reverse engineering scenarios where AI hallucinations are common. **Key features:** - Reverify proposes claims; its tools check against ground truth. - It verifies every claim against the actual artifact, ensuring only verified facts are accepted. - It uses a deterministic Python toolkit, making it easy to integrate with existing systems. - It works seamlessly with various languages and platforms, including Claude Code, Codex, Gemini CLI, and OpenCode. **Tech stack:** Pure Python toolkit. **Why it matters:** Reverify is crucial in scenarios where AI models are prone to hallucinating, such as reverse engineering. By ensuring that every claim is checked against reality, Reverify helps maintain the integrity and reliability of AI-driven reverse engineering efforts. **Repository:** 2akouwu/reverify **Primary language:** Python **GitHub stars:** 1312 **GitHub description:** Stop your AI from making things up — it proposes, deterministic tools decide, every claim checked against ground truth with evidence. Grounded facts and context survive resets. Reverse engineering is the proving ground. MCP server + CLI.

### Overview: Overstory is a multi-agent orchestration system designed for AI coding agents. It allows you to spawn and manage multiple Claude Code agents in isolated git worktrees, coordinating them through a custom SQLite mail system, and merging their work back with tiered conflict resolution. The system supports pluggable runtimes, including Claude Code, Pi, and Gemini CLI, among others. ### Key Features: - **Multi-Agent Orchestration**: Spawns worker agents in isolated git worktrees, coordinating them through a custom SQLite mail system. - **Custom Mail System**: Uses a custom SQLite mail system for communication between agents. - **Tiered Conflict Resolution**: Merges work back with tiered conflict resolution to handle compounding error rates and debugging complexity. - **Pluggable Runtimes**: Supports 11 runtimes, including Claude Code, Pi, Gemini CLI, Aider, Goose, Amp, and your own adapter. - **Headless Agents**: Spawns Claude agents headless by default, allowing the UI to surface them through a web interface. - **Opt-In Live Steering**: Provides an opt-in escape hatch for live steering using `tmux attach`. ### Tech Stack: - TypeScript - Custom SQLite mail system - Pluggable runtimes ### Why It Matters: - **Multi-Agent Orchestration**: Empowers teams to work more efficiently by managing multiple agents simultaneously. - **Customization**: Allows for easy swapping of runtimes, making Overstory highly flexible and adaptable. - **Comprehensive Error Handling**: Handles compounding error rates and debugging complexity effectively.

### Overview: Detoxify is a comprehensive project designed to predict toxic comments across three Jigsaw Toxic Comment Challenges. The model and code are built using Pytorch Lightning and Hugging Face Transformers, offering improved performance and consistency across different datasets and languages. ### Key Features: - **Improved Models**: The project features updated models for the `original`, `unbiased`, and `multilingual` challenges, achieving higher AUC scores. - **Lightweight Models**: Smaller models trained with Albert are available for inference, providing faster processing times. - **Community Contribution**: The project is maintained by Laura Hanu and James Thewlis at Unitary, aiming to combat harmful online content. ### Tech Stack: - **Dependencies**: - 🤗 Transformers - ⚡ Pytorch Lightning ### Why it Matters: Detoxify is a significant contribution to the field of natural language processing and machine learning, particularly in the area of toxic comment classification. By providing robust and consistent models across different challenges and datasets, it helps in improving the detection and reduction of harmful content online.

### Overview: Caliber is an open-source context layer designed to seamlessly integrate with various coding agents like Claude Code, Cursor, and Codex. It ensures that your agents have up-to-date and contextually relevant information without the need for manual updates. ### Key Features: - **Contextual Updates:** Automatically syncs agent configurations with the latest changes in the repository. - **Skill and Rule Management:** Writes a skill once, and it becomes available across all agents. - **Lean Configuration:** Ensures that only the necessary files are maintained, reducing redundancy. ### Tech Stack: TypeScript ### Why It Matters: Caliber simplifies the process of managing and updating agent configurations, ensuring that your agents always have the most current and relevant information. This tool is essential for maintaining a cohesive and efficient development environment across multiple agents and platforms. ### Repository: **Repository:** caliber-ai-org/ai-setup **GitHub Stars:** 1309 ### Primary Language: TypeScript ### GitHub Description: Continuously sync your AI setups with one command. Codebase tailor suited agent skills, MCPs and config files for Claude Code, Cursor, and Codex. ### GitHub Description (Excerpt): <p align="center"> <img src="assets/readme/logo.png" alt="Caliber" width="120"> </p> <p align="center"><b>The open-source context layer for coding agents.</b><br> Claude Code · Cursor · Codex · OpenCode · GitHub Copilot</p> <p align="center"> <a href="https://

### Overview: Hopsworks is a Real-Time AI Lakehouse designed for ML with a Python-centric Feature Store. It provides a secure, governed platform for developing, managing, and sharing ML assets, including features, models, training data, batch scoring data, and logs. ### Key Features: - **Real-Time AI Lakehouse**: Supports real-time data processing and analytics. - **Python-centric Feature Store**: Enables efficient management and serving of features. - **Modular Platform**: Can be used as a standalone Feature Store, managed ML model server, or feature pipeline and training pipeline manager. ### Tech Stack: Java ### Why it Matters: Hopsworks is a powerful tool for modern data-intensive AI and ML projects, offering a secure, governed platform for developing, managing, and sharing ML assets, features, models, training data, and batch scoring data.

Overview: MCPJungle is a self-hosted MCP gateway designed to manage multiple MCP servers efficiently, providing a single control point for developers and teams. Key features: - Centralized MCP server management - Unified discovery and access control - Shared infrastructure for team collaboration - Local setup for personal use Tech stack: Go Why it matters: MCPJungle simplifies the management of multiple MCP servers, reducing clutter and improving consistency across teams. Repository: mcpjungle/MCPJungle GitHub stars: 1310 GitHub description: One place to manage & connect to all your MCP servers